Paths, durativity, and salient goals

In this section I examine distributional differences that exist even when -ni and -made combine with path verbs, where their distribution is usually assumed to be equivalent. These differences indicate a much more fundamental distinction between them as regards which participant they each mark in a given motion description. Consider (6), where -made (but not -ni) is unacceptable with the verb hairu ‘enter’ (cf. Tsujimura Reference Tsujimura2002).

1. (6) [John is just outside the tower.] John-wa too-ni/*made haitta. John-top  tower-to/until went. in‘John went into the tower.’

The difference between hairu and the verbs in (1a) is that the latter are accomplishments and hairu is an achievement (see Kindaichi (Reference Kindaichi and Kindaichi1976) and Tsujimura (Reference Tsujimura1996, Reference Tsujimura and McClure2003) on the Vendler aspectual classes in Japanese). Accomplishments are telic and necessarily durative, whereas achievements are telic and punctual. At first blush it appears that -made only occurs in durative predicates (as Kuno Reference Kuno1973 suggests). However, there is evidence that this is not quite correct, and that -made instead occurs in motion constructions in which the path of motion is complex, a slightly different notion than durativity. Path complexity and durativity are of course not unrelated. The aspectual behavior of motion predicates is often analyzed in terms of a homomorphic mapping between paths and events that preserves boundedness and/or some degree of mereological complexity (see in particular Wechsler Reference Wechsler2001, Reference Wechsler, Beermann and Hellan2003, Reference Wechsler, Erteschik-Shir and Rapoport2005 and Beavers Reference Beavers2002, Reference Beavers2006, Reference Beavers, Dölling, Heyde-Zybatow and Schäfer2008; see also Dowty Reference Dowty1991; Tenny Reference Tenny, Sag and Szabolcsi1992, Reference Tenny1994; Jackendoff Reference Jackendoff1996; Krifka Reference Krifka and Rothstein1998; Hay, Kennedy, & Levin Reference Hay, Kennedy and Levin1999). Following Beavers (Reference Beavers, Dölling, Heyde-Zybatow and Schäfer2008), the punctuality of predicates such as (6) correlates with the fact that events of entering typically involve transitional paths, defined as two-place paths consisting only of initial and final points. This is distinct from paths of durative motion events (e.g. crossing a bridge), which include initial and final points as well as middle portions of arbitrary mereological complexity. The contrast is exemplified by the minimal pair in (7).

1. (7)

   1. (a) The fugitive crossed the border. (punctual)

   2. (b) The fugitive crossed the desert. (durative)

In (7a) the path is inherently transitional (one side of the border to the other side) while in (7b) it is necessarily complex (one end of the desert to the other via the intervening space). Correspondingly, (7a) is punctual and (7b) is durative.Footnote ⁷ Crucially, we can reconstrue the event in (6) such that the path is complex, in which case for many speakers the accepability of -made improves considerably.

1. (8) [John is outside a wall around the perimeter of the tower's area, a short walk from the tower door, and his goal is the chamber inside the tower.] John-wa too-ni/?made haitta. John-top tower-to/until went. in‘John went in as far as the tower’

Thus the length of the path has an effect on -made's acceptability. However, due to the close correlation of path complexity and durativity, it is difficult to tell which factor conditions the acceptability of -made (if it is possible at all to tease these apart). In fact, there are data suggesting that -made is sensitive only to path complexity. In particular, there are certain motion events that, due to the peculiarity of the type of motion involved, are punctual but are still compatible with -made, provided the path of motion is complex. An example is given in (9).Footnote ⁸

1. (9)

   1. (a) [In the context of a digital display going from 5 to 500 in a single change.]Denshikei-ga 500-made/ni agatta. electronic-degree-measure-nom 500-until/to went.up‘The display went up to 500.’

   2. (b) [In the context of a digital display going from 10 to 11, the maximum.] Denshikei-ga 11-ni/?#made agatta. electronic-degree-measure-nom 11-to/until went. up‘The display went up to 11.’

In (9a), a digital readout ‘moves’ over 495 possible readings to a final reading of 500, but does so instantaneously. Thus the event is punctual but the path is still complex. That it is traversed instantly is due to the peculiarity of the type of movement described, which serves to collapse or eliminate the homomorphic relationship between the event and the path. However, if the path complexity is somehow compromised, the acceptability of -made goes down, as in (9b). Thus -made is sensitive to path complexity rather than durativity.

Furthermore, in some cases the acceptability of -made is dependent not on the existence of intermediate locations between the source and goal but instead on the existence of additional locations on the path after the goal, as in (10):

1. (10)

   1. (a) [John is on the first floor of a two-story building.] John-ga ni-kai-ni/?#made nobotta.John-nom 2nd-cl.floor-to/until went.up‘John went up to the second floor.’

   2. (b) [John is on the first floor of a ten-story building.] John-ga ni-kai-ni/made nobotta.John-nom 2nd-cl.floor-to/until went.up‘John went up to the second floor.’

Although not entirely unacceptable to many speakers, in the context of (10a) -made is significantly less natural than -ni, but becomes more natural if there are additional floors beyond the second floor, as in (10b). A similar distinction exists in English:

1. (11)

   1. (a) [In a two-story building]

      John went up to/?#until/?#as far as the second floor.

   2. (b) [In a ten-story building]

      John went up to/until/as far as the second floor.

English until/as far as sound better in contexts with a complex path even if some parts of the path play no role in the event, whereas up to is always acceptable. This indicates that -made is sensitive to the complexity of the path after the source point, independent of the path's role in the event. The fact that this often correlates with durativity follows from its homomorphic relationship to the event plus the fact that quite often the ‘extra’ parts of the path come before the goal and thus are implicated in the event.Footnote ⁹ The postposition -ni is not sensitive to either factor, as seen in all of the data above, where -ni is always acceptable.Footnote ¹⁰

Conversely, there are cases where -made but not -ni is acceptable with path verbs, namely when the goal is not the one that is most natural for a given verb in a given context. For expository purposes I call this the ‘salient goal effect’. This distinction is conditioned by the particular kind of path verb being used. Muehleisen and Imai (Reference Muehleisen, Imai, Lee, Sweetser and Vespoor1996) and Tsujimura (to appear) argue that Japanese path verbs can be divided into two classes depending on which aspects of the path are directly encoded by the verb. The first are ground-path verbs, which incorporate constraints on the physical geometry of the ground (the overall path) into their meanings (e.g. koeru ‘cross’, wataru ‘go over’, kuguru ‘pass through’, hairu ‘enter’). Of course, any constraints a verb imposes on the ground/path inherently place constraints on the goal, which must be compatible with the geometry of the path.Footnote ¹¹ Crucially, if the figure does not traverse the entire path described by the verb, but instead only traverses some subportion of it, then only -made but not -ni is acceptable (data similar to this are also discussed by Matsumoto Reference Matsumoto, Tanaka and Matsumoto1997: 187):Footnote ¹²

1. (12)

   1. (a) [John is at one end of a series of bridges.] John-wa hitotume-no hashi-no mukoo gawa-ni/?made watatta.John-top first-gen bridge-gen yonder side-to/until went.over‘John crossed/went over the first bridge.’

   2. (b) [John is at one end of a bridge.] John-wa hashi-no tochuu-made/*ni watat-te hikikaeshita.John-top bridge-gen halfway-until/to went.over-prt return‘Going to halfway along the bridge, John returned.’

Thus -ni in conjunction with a ground-path verb must mark a goal that matches the intrinsic path geometry of the verb (in the case of (12a) the other side of the first bridge). Note that the acceptability of -made in (12b) is significantly helped by the supporting context that the agent only went so far on the bridge before returning, perhaps out of fear. This effect is much like the imperfective paradox of Dowty (Reference Dowty1979: 133ff.), where in a sentence such as We were flying to Paris when our plane was forced to land in New York the underlying predicate refers to a motion event in some possible world of going to Paris, even though in the real world this event did not necessarily occur and the actual goal was instead New York. In this case we have two goals, one implicated by the verb and one actually arrived at, just as in (12b). In section 4 I return to this connection. The important point for now is that -ni is unacceptable unless it marks a goal satisfying the constraints imposed lexically by the verb itself, independent of the actual event.

The second class of verbs identified by Muehleisen & Imai (Reference Muehleisen, Imai, Lee, Sweetser and Vespoor1996) and Tsujimura (to appear) are direction-path verbs, which incorporate some notion of directionality into their meaning (e.g. noboru ‘climb’ or oriru ‘go down’). Interestingly, while these verbs do not encode explicit geometry, some of them may nonetheless show salient goal effects similar to ground-path verbs when certain contextual factors place a restriction on possible goals. For example, consider the following in the context of John being at the bottom of a 20-story tower:

1. (13)

   1. (a) [John intends to go to the 10th floor.] John-wa juk-kai-ni/made nobotta/agatta.John-top 10th-cl.floor-to/until went.up/went.up‘John went up to the 10th floor.’

   2. (b) [John intends to go to the top.] John-wa juk-kai-made/?#ni nobotta/agatta.John-top 10th-cl.floor-until/to went.up/went.up‘John went up until the 10th floor.’

Here the figure has certain intentions about the intended goal in the event. When the actual goal is the contextually salient (intended) one, as in (13a), then both -ni and -made are acceptable. When the actual goal is not the salient one, as in (13b), then for many of my informants -made is the only natural way to mark the actual goal. Furthermore, for speakers who get the contrast in (13b), there is one subclass of direction-path verbs for which any type of salient goal effect is largely absent, namely the deictic motion verbs iku ‘go’ and kuru ‘come’:

1. (14) [John is at one end of a bridge and intends to go across the entire bridge.] John-wa hashi-no tochuu-made/ni itta/kita.John-top bridge-gen middle-until/to went/came‘John went/came until halfway across the bridge.’

Here no context favors one adposition over the other, even in the same contexts that figured in other salient goal effects.

However, not all speakers share the judgments in (13b). One anonymous reviewer finds both -ni and -made acceptable in (13b) in the given context. I also note that the classification of noboru and agaru ‘go up’ is not clear-cut: although Tsujimura (to appear) classifies them as direction-path verbs, Muehleisen & Imai (Reference Muehleisen, Imai, Lee, Sweetser and Vespoor1996) classify them as ground-path verbs. This discrepancy is not surprising; English climb likewise has several meanings, including ‘go upwards’, ‘clamber (in any direction)’, and ‘completely ascend’. Indeed, the same reviewer who rejects the distinction in (13b) also gives the judgments in (15), demonstrating a ground-path salient goal effect:

1. (15) Hashigo-no tochuu-made/*ni nobotta/agatta.ladder-GEN middle-until/to went.up/went.up‘I climbed up to the middle of the ladder.’

Whether the effect in (13) is found with all speakers or not does not, however, affect my overall conclusion that -ni and -made are distinct regarding salient goals, since it is sufficient to show that there is some distinction between them with some path verbs, as (12) already demonstrates.

In summary, it appears that with certain path verbs the acceptability of -ni (but not -made) is dependent on either (i) a certain path geometry inherent in the verb and/or possibly (ii) contextual factors (of which intentionality appears to be one). Each factor determines a certain set of appropriate (‘salient’) goals, and -ni may not mark any goal not in this set. I do not propose that these two factors are the only ones that may determine when -ni is unacceptable. What is crucial is simply that even among path verbs -ni and -made are not entirely interchangeable: -made is subject to certain constraints on the complexity of the path, while -ni is subject to certain constraints on goal salience.

Event-dependent and verb-dependent restrictions

So far we have seen the following distributional restrictions:Footnote ¹³

1. (16)

   

In this section I will reduce these restrictions to one simple contrast: -made realizes participants in motion events, while -ni realizes arguments of motion verbs. I first clarify the distinction I make between motion events and motion verbs (or, more properly, motion predicates). Any given situation in the real world has a range of properties that a specific event-denoting lexeme may or may not encode. For example, an event of John running into the house necessarily involves both a manner and a goal, but three different descriptions of the event might highlight different properties and remain neutral about others. The sentence John went into the house is neutral with respect to manner but imposes a constraint on the goal (its topography and directionality from the deictic center), while John ran is neutral with respect to the goal, but imposes a constraint on the manner (running). John ran into the house encodes both components explicitly.

Distinguishing properties of events from those encoded by the verb, we can classify each restriction in (16) in terms of whether it depends on the event being described or the verb being used. First, the verb-type restriction on -ni is clearly a verb-based restriction, since -ni may only realize the goal of a particular motion event if the verb is a path verb, as shown again in (17) for two different descriptions of an event of John crawling to the station.

1. (17)

   1. (a) John-wa (*eki-ni) hatta.John-top station-to crawled‘John crawled to the station.’

   2. (b) John-wa (hatte) eki-ni itta.John-top crawling station-to went‘John went to the station (crawling).’

This suggests that -ni is restricted to occur only with verbs that encode goals explicitly, and not simply in descriptions of events that have goals. Alternatively, one could suppose that -ni is somehow restricted to descriptions of motion events that have goals, but that hau ‘crawl’ and other manner verbs specifically deny the existence of a goal. Thus the combination of -ni with a manner verb yields a contradiction. But this position is clearly not tenable since manner verbs with -made are compatible with goal interpretations, as seen again in (18):

1. (18) John-wa eki-made hatta.John-top station-until crawled‘John crawled to the station.’

Thus manner verbs are neutral with regard to goal interpretations (though see section 5 for some variation on this) and -ni is restricted to occur only with path verbs. The salient goal constraint of -ni in (16) is likewise a restriction based on the verb rather than the event. This may seem counterintuitive, since this effect seems to be at least partly conditioned by context (i.e. properties of specific events), as the data in (13) show. But as seen in (14), the primary conditioning factor is the verb class: the salient goal effect only shows up with ground-path and (depending on the speaker) non-deictic direction-path verbs, but not deictic motion verbs. Thus it is only with respect to appropriate verb classes that the role of context comes into play for determining exactly what the salient goal is, showing that this is first and foremost a verb-based restriction.

Turning to the path complexity constraint of -made, this is a purely event-dependent property. This should be obvious from the discussion in the previous section. Verbs like agaru, noboru ‘go up’, and even hairu ‘enter’ are all by and large neutral as regards the complexity of the path. The acceptability of -made with any of these verbs is conditioned by the complexity of the path as determined by context. This would suggest that any sensitivity -made has to path complexity is due entirely to whether or not the motion event being described has a complex path, regardless of the verb. Collapsing (16) into two categories, we get a simpler picture of the distributional restrictions on -ni and -made.

1. (19)

   

Thus, while -ni is restricted to certain classes of motion verbs regardless of the event, -made is restricted to certain motion events regardless of the verb class. In the next section I present additional data that further validates this distinction but also allows us to generalize it by eliminating the specifically motion-based component.

Argument-marking vs. participant-delimitation

Both -ni and -made have uses outside of motion constructions, and here the verb-dependent vs. event-dependent contrast is further supported, without reference to motion at all. In general, -ni is often analyzed as an argument-marker, in particular as a dative case (Kuno Reference Kuno1973, Martin Reference Martin1975, Tsujimura Reference Tsujimura1996). Some of its additional uses are shown in (20) (glosses are from original sources; for a more exhaustive catalogue see Sadakane & Koizumi Reference Sadakane and Koizumi1995, based on Martin Reference Martin1975).Footnote ¹⁴

1. (20)

   1. (a) Mary-ga boku-ni kono hon-o kureta.Mary-nom I-dat this book-acc gave‘Mary gave me this book.’ (dative/recipient)

   2. (b) John-wa Mary-ni hon-o yom-ase-ru.John-top Mary-dat book-acc read-cause-non.pst‘John will make Mary read a book.’ (causee)

   3. (c) Teeburu-no-ue-ni koppu-ga aru.table-gen-top-at cup-nom exists‘There are cups on the table.’ (location in existential) (Kuno Reference Kuno1973: 127–139, 352, exx. (3a), (7), (5a))

   4. (d) E-ga doroboo-ni nusum-are-ta.painting-nom thief-by steal-pass-pst‘The painting was stolen by the thief.’ (logical subject of passive) (Tsujimura Reference Tsujimura1996: 233, ex. (169))

Thus the role of the -ni-marked participant is always determined by the verb, with no unifying semantics common to all such participants that is inherent to -ni (just as nominative/accusative cases carry no semantics).Footnote ¹⁵ Rather, we can explain the distribution of -ni in motion constructions in terms of verb class. While all motion verbs entail the existence of a path, manner verbs take only an event argument while path verbs take a goal argument and an event argument, and dative is how goals are marked in Japanese.Footnote ¹⁶ (I assume for expository purposes that external arguments are introduced by functional heads; Kratzer Reference Kratzer, Rooryck and Zaring1996.) Sample meaning representations for aruku ‘walk’ and noboru ‘go up’ are given in (21), where e is a motion event in the domain of events U _E, p is its path in the domain of directed paths U _D, and y∊U _D is the final part of p at which the figure arrives at the end of e.Footnote ¹⁷

1. (21)

   1. (a) [[aruku]]≔λe∃p[walk(e, p)]

   2. (b) [[noboru]]≔λyλe∃p[go.up(e, y, p)]

By contrast, -made serves a range of delimitation functions as shown in (22). In such uses Kuno (Reference Kuno1973: 108) defines -made as ‘continuously until/to X’, and notes that such examples are always durative. In general these uses of -made are similar to its uses in motion constructions: in all cases it delimits some participant in the event/state and imposes some type of complexity constraint.Footnote ¹⁸

1. (22)

   1. (a) Ohiru-made kore-o shite-kudasai.noon-until this-acc do-please‘Please do this until noon.’ (temporal)

   2. (b) Yuka-kara yane-made nan-meetoru arimasu ka?floor-from roof-until how-many-meters are ques‘How many meters from the floor to the roof?’ (spatial numeral) (Kuno Reference Kuno1973: 108–110, exx. (1a), (6))

   3. (c) Kono hooru-wa nisen-nin-made haireru.this hall-top 2,000-cl.people-until hold‘This hall can hold up to 2,000 people.’ (non-spatial numeral)

   4. (d) Hikooki-ga deru-made tomodachi-to hanashite ita.plane-nom leave-until friend-with talking was‘Until the plane left I was talking with my friend.’ (propositional) (Makino & Tsutsui 1986: 226–228)

Thus the inherent semantics of -made is that it delimits some participant in the event/state (a temporal trace, a path, a numerical range, etc.). It is a matter of pragmatic or lexical semantic inference, based on -made's complement and the event/state being described, which participant it delimits. I sketch a model of this here (following the mereological event semantics of Krifka Reference Krifka and Rothstein1998). We can define a limit as in (23), where g is the limit of some x in event e iff x is a participant in e (i.e. there is a Parsons-style (1990) θ-relation between x and e) and g is the final part of x (i.e. it precedes no other part of x).Footnote ¹⁹ This definition places no constraints on the domain of x and g. They may be members of any domain (e.g. the domain U _D of directed paths or the domain U _T of times). However, since g is a subpart (⩽) of x they must be in the same domain, and furthermore this domain must be one that has a precedence function ≪ defined for it such that one part of x can be said to be ‘before’ or ‘after’ another (specifically g is the ‘last’ part of x).Footnote ²⁰

1. (23) ∀e∀g∀x[LIMIT(e, g, x)↔∃θ[θ(e, x)∧g⩽x∧¬∃x″[x″⩽x∧g≪x″]]]

   (‘g is the limit of x in e iff x is a participant in e and g is a part of x and g precedes no other part of x.’)

I define -made as in (24), where -made takes as its arguments some participant g (i.e. its complement) and the event-denoting predicate P (i.e. the VP it modifies), and says that g is the limit of some participant x in event e.

1. (24) [[-made]] (Preliminary)≔λgλPλe∃x[P(e)∧LIMIT(e, g, x)]

   (‘g is the limit of x in event e described by P.’)

By (23) the participant x -made delimits must be θ-related to the event and from the same domain as g. But since it is existentially bound its interpretation is determined purely by contextually-based inference, giving rise to multiple possible interpretations depending on the event. For example, when -made's complement is a temporal DP or CP (g∊U _T) as in (22a), x must be the temporal trace t of e (the only participant in e that is in U _T).Footnote ²¹ This use of -made is available for all predicates, since all events have a temporal trace, even motion events, as shown in (25) (ignoring tense; Fig is the figure thematic role, j is John):

1. (25) John-ga ohiru-made aruita.John-nom noon-until walked‘John ran until noon.’ ∃e∃x[Fig(e, j)∧∃p[walk(e, p)]∧LIMIT(e, noon, x)] (x=t)

The combination of spatial DPs (g∊U _D) and motion events is more complex. Applying (24) to the representations in (21) yields (26) for complement DP X referring to g (assuming that noboru's null goal argument is existentially bound):

1. (26)

   1. (a) [[X-made aruku]]=λe∃x[∃p[walk(e, p)]∧LIMIT(e, g, x)]

   2. (b) [[X-made noboru]]=λe∃x[∃y∃p[go.up(e, y, p)] ∧ LIMIT(e, g, x)]

All motion events involve a path p which is eligible to be delimited spatially. For manner verbs this is ostensibly the only appropriate participant, and thus in (26a) x must be p, resulting in an interpretation much like a path verb (cp. (26a) to (21b)), though I discuss an exception to this below. For path verbs, as in (26b), there are more possibilities. The obvious interpretation is again that x=p. However, path verbs also select a goal argument y which must be reconciled with the limit g imposed by -made. The only non-contradictory resolution is that the two are equated (y=g), making it seem that -made is a goal-marker like -ni.

However, recall the salient goal effect, where -made may mark a goal not selected for by the verb. This would be a case where putatively y and g in (26b) are not the same. As it stands this is a contradiction. But, as discussed in section 3.1, we could adopt here an analysis analogous to that of the imperfective paradox. Dowty (Reference Dowty1979: 149, ex. (25)) analyzes the imperfective paradox via an intensional operator PROG, where (informally) PROG φ is true at some interval of time iff φ happens to be true at some appropriate larger interval of time in all ‘inertia worlds’ (worlds minimally different from the real world in relevant ways). A complete discussion of this is far beyond the scope of this paper. However, I tentatively assume that there is some roughly equivalent operator M that relates an event e to a predicate P such that M(e,P) is true iff P is true at some appropriate superevent of e in all inertia worlds. A revised analysis of -made is given in (27). This produces representations such as (28) for (15) (repeated here), where noboru inherently describes an event of going to the top of the path (I is the implicit subject):

1. (27) [[-made]] (Rev 2)≔λgλPλe∃x[M(e, P) ∧ LIMIT(e, g, x)]

   (‘g is the limit of x in e, M-related to P.’)

2. (28) Hashigo-no tochuu-made/*ni nobotta.ladder-GEN middle-until/to go.up‘I climbed up to the middle of the ladder.’ ∃e∃x[Fig(e, I)∧M(e, λe∃p[go.up(e, top, p)])∧LIMIT(e, middle, x)]

The limit marked by -made in (28) is interpreted as the point at which the figure actually arrives (the middle of the ladder), despite the restriction that in every inertia world the goal is the top. If the goal were realized as tochuu-ni it would be a direct argument of noboru and would thus clash with its lexically-defined constraints that the goal be the top, ruling out -ni.

Of course, as discussed in section 3 salient goal effects only occur with ground-path and possibly non-deictic direction-path verbs. To capture this I assume that (perhaps due to some Gricean conversational maxim) the intensional effects of M only apply when there is some explicit mismatch between the verb's selectional restrictions and those of -made that would otherwise yield a contradiction. Only ground-path and non-deictic direction-path verbs impose (lexical or contextual) selectional restrictions on their goal arguments, thus giving rise to the possibility of a contrast with what is marked by -made. Deictic motion verbs do not impose selectional restrictions on their goal arguments (and for some speakers neither do the rest of the direction-path verbs), and manner verbs do not even select for goal arguments, so that neither verb class gives rise to salient goal effects.

There is also an interesting third possibility inherent in (26b). Since the verb's goal argument y is a member of U _D, it is also eligible to be delimited by -made, just like the path p. Of course, delimiting y delimits p, so that there should be no semantic difference between x=y and x=p. However, there is a noteworthy case where it is sometimes possible to get -ni and -made in the same sentence:

1. (29)

   1. (a) Gakkoo-ni-wa genkan-made-sika hair-anakat-ta.school-to-top entrance way-until-only enter-neg-pst‘(He) entered the school only as far as the entrance way.’

   2. (b) Sono ana-ni oku-made te-o tukkonda.this hole-to deep-until arm-acc stuck‘(I) put my arm deep into the hole.’ (Tsuguro Nakamura, p.c.)

Crucially, such sentences are only acceptable when -ni describes the goal selected for by the verb and -made adds additional, more event-specific information about this goal, resulting in a case where -made delimits the goal y (i.e. x=y in (26b)) in a meaningful way, rather than delimiting the path p. That this is a possibility is an interesting prediction of this approach, and represents another way that -ni and -made can have different functions even when they cooccur.

Furthermore, even non-motion verbs may take spatial limits, provided the event is contextually construable as involving motion, as in (30) in the context of riding a train (Matsumoto Reference Matsumoto, Tanaka and Matsumoto1997: 188 makes a similar point citing similar data):

1. (30)

   1. (a) Nagayo-made bentoo-o tabeta.Nagayo-until meal-acc ate‘(I) had a meal all the way to Nagoya.’

   2. (b) Kyoto-made neta.Kyoto-until slept‘(I) slept all the way to Kyoto.’ (Tsuguro Nakamura, p.c.)

Although taberu ‘eat’ and neru ‘sleep’ are not motion verbs, in a context where the eating and sleeping activities occur on a train, -made is acceptable with a spatial limit on a goal interpretation.Footnote ²² For simplicity's sake we can assume that the predicates in (30) are coerced to interpretations that involve motion and thus involve a new path participant p (as well as a figure and possibly other participants), in which case the definition in (24) can apply to them with x=p.

Matsumoto's (Reference Matsumoto, Tanaka and Matsumoto1997) data in (5) (repeated here) shows a similar effect, wherein there is a reading that Taroo was swimming in the swimming pool on a cruise liner heading to the port. However, this predicate is already a motion predicate, so that there is also the simple reading that Taroo swam to the port.

1. (31) Taroo-wa minato-made sanzyu-pun-kan/de oyoida.Taroo-top port-until 30-minutes-for/in swam‘Taroo swam to the port in 30 minutes.’ (goal-marking -made) ‘Taroo swam for 30 minutes till he got to the port.’  (temporal -made)

This represents a case of -made with a manner verb where the entity x delimited by -made is not equated with the path p of the verb (an exception to the discussion above). However, assuming that coercion of the sort discussed for (30) is possible even for verbs that already describe motion, we can explain the ambiguity of (31) in terms of the availability of two path participants that might potentially be delimited: that of the swimming event and that of the coerced cruise liner event. Thus the two readings of (31) follow from a single definition of -made, capturing Matsumoto's original insight without assuming separate goal-marking and temporal uses.

Still further, -made can also mark limits of numerical values as in (22b) and (22c). A full account of how measurements are analyzed is again beyond the scope of this paper. However, we could naïvely assume that such stative events have measurement participants that are continuous numerical ranges on the real number line ordered by standard numerical precedence relations such as ‘less than (or equal)’. This makes measurements eligible for having a limit imposed on them, giving rise to another use of -made having nothing to do with either motion or temporality. Thus a single definition allows a range of interpretations for different sorts of events, provided only that the given event participant is compatible with having a limit of the type of the complement of -made.

Finally, as noted in section 3, for many speakers -made imposes a constraint that the participant it delimits is complex. In the case of temporal traces and paths this in turn often determines durativity of the event, where the homomorphic θ-relation relating each participant back to the event preserves mereological complexity (Beavers Reference Beavers, Dölling, Heyde-Zybatow and Schäfer2008). However, recall also from section 3 that the complexity constraint is technically independent of the event: the relevant additional subpart could occur on the path before or after the goal. To capture this we could modify (27) so that x is an initial (INI) subpart of a larger event-independent entity x′ (i.e. x⩽x′ and x and x′ share the same initial point; Krifka Reference Krifka and Rothstein1998: 206, ex. (36a)), where g is the final point on x in e as in (27) and x′ must be a complex object (CO) (i.e. having at least three subparts; Beavers Reference Beavers, Dölling, Heyde-Zybatow and Schäfer2008: 254, ex. (24b)):

1. (32) [[-made]] (Final)≔λgλPλe∃x∃x′[M(e, P)∧LIMIT(e, g, x)∧INI(x, x′)∧CO(x′)]

   (‘g is the limit of x (initial subpart of complex x′) in e, M-related to P.’)

The extra part of x′ other than x may lie before or after g on x′. For example, (33) represents the analysis of (10b), where John traverses a portion of a larger path that has multiple possible endpoints after the goal. Here x is the path from the first to the second floor, while x′ is a complex superpath from the first to the tenth floor. Alternatively, the sentence could occur in the context that x=x′ (presumably the default interpretation), in which case x would have to be complex.

1. (33) [John is on the first floor of a ten-story building.] John-ga ni-kai-made nobotta.John-nom 2nd-cl.floor-until went.up‘John went up to the second floor.’ ∃e∃x∃x′[Fig(e, j)∧M(e, λe∃y∃p[go.up(e, y, p)])∧LIMIT(e, 2nd.floor, x)∧INI(x, x′)∧CO(x′)]

While this analysis is only a sketch, it is clear that we can infer how the bound is interpreted from the type of bound -made marks and the type of event it modifies. This gives rise to multiple possible interpretations dependent on context, even with a single predicate, something not possible with -ni. Thus this analysis captures the fact that despite the apparent overlap of -ni and -made in motion constructions, they are in fact realizing two entirely different things: one is truly picking up on what happens in the event (the real goal), the other on what the verb highlights about the event (the salient/expected goal). It is only due to the fact that in most cases these two goals are exactly identical that -ni and -made appear to have the same function.

Morphosyntactic evidence

A final question to be asked is whether there is any morphosyntactic correlate to this semantic distinction. Intuitively, the cases where -ni and -made are interchangeable reflect the difference between a syntactic argument and an adjunct. As it turns out, it is surprisingly hard to find evidence for this contrast. First, in motion contexts -ni appears to be an oblique postposition rather than a case-marking postposition, insofar as this can be determined in Japanese. For example, nominative -ga, accusative -o, and recipient -ni are all deletable in some contexts, including clefting constructions and under topic-marking by the topic-marker -wa (Kuno Reference Kuno1973; Miyagawa Reference Miyagawa1989; Dubinsky Reference Dubinsky, Postal and Joseph1990, Reference Dubinsky1994; Sadakane & Koizumi Reference Sadakane and Koizumi1995). Oblique postpositions such as -de ‘at, with’, -kara ‘from’, and -made ‘until’ are never deletable in these contexts (see e.g. Kuno Reference Kuno1973: 357).Footnote ²³ In motion constructions, -ni patterns like obliques in that it is not deletable, as shown in (34) under topic-marking by -wa, contra nominative -ga:

1. (34)

   1. (a) John-ga eki-ni itta.John-nom station-to went‘John went to the station.’ (no topic marking by -wa)

   2. (b) John-wa eki-ni itta.John-top station-to went‘As for John, he went to the station.’ (-ga deleted under topic -wa)

   3. (c) Eki-ni-wa/*eki-wa John-ga itta.station-to-top/station-top John-nom went‘As for the station, John went to it.’ (-ni not deletable under topic -wa Footnote ²⁴)

Thus -ni here is oblique, so that at best a -ni-marked goal is an oblique argument.

Other tests for argumenthood vs. adjuncthood are likewise of no use. Japanese allows scrambling, so word order tests (e.g. adjuncts occur further from the head than arguments) do not necessarily apply. Optionality also does not help, since Japanese allows massive amounts of pro-drop (nearly any DP or PP is optional). Iterativity (the ability of adjuncts but not arguments to be iterated, e.g. The beautiful green vegetables vs. *Sandy saw Kim a dog; Schütze Reference Schütze1995: 102f.) is of very limited use given the nature of the semantic domain: a motion event may have at most one goal, so it makes little sense to look for multiple goal-phrases per clause. However, with this in mind, recall the data in (29), wherein both -ni and -made can cooccur provided -ni describes the goal selected for by the verb and -made adds additional, more event-specific information about this goal. This certainly suggests that at least one of these postpositions (presumably -made) marks an adjunct. However, without further corroborating evidence it is difficult to tell what grammatical function each PP has, though this is not to say that a syntactic contrast does not exist. Nonetheless, what is crucial is that there is a semantic contrast, as discussed above, demonstrating significant differences between these two postpositions.